Доменно-адаптивна методологія багатокритеріальної оцінки якості масштабованих растрових зображень

Abstract

This paper presents a domain-adaptive methodology for evaluating the quality of scaled raster images, bridging the evaluation gap between classical mathematical interpolation and modern AI-based Super-Resolution (SR). Highlighting the fundamental limitations of pixel-based metrics (MSE, PSNR) caused by the perception-distortion tradeoff, the study introduces a structured trimetric framework (PSNR, SSIM, LPIPS). This approach establishes a unified and objective standard for benchmarking diverse upscaling algorithms